Modular machine learning systems, databases, methods, and computer program products for developing and deploying automated image segmentation programs

Abstract

A system for training and deployment of automated image segmentation algorithms is provided. The system includes a database; one or more user interfaces for curating collections of images and annotating images that are stored in the database; a structured library of annotations that form a dictionary for defining features within an image; a means for training a deep learning model according to a set of annotations applied to a collection of images; a means for deploying a plurality of deep learning models from the set of applied annotations, the plurality of deep learning models including a nested set of annotations such that a first model segments a first set of features within the image and a second model segments the first set of features of the first model and at least one additional feature of the image; and a means for transfer learning, wherein at least one of the plurality of deep learning models deployed for a first collection of images is used to accelerate retraining of the deep learning model for a second class of images defined by the second collection.

That which is claimed is: 
     
         1 . A system for training and deployment of automated image segmentation algorithms, the system comprising:
 a database;   one or more user interfaces for curating collections of images and annotating images that are stored in the database;   a structured library of annotations that form a dictionary for defining features within an image;   a means for training a deep learning model according to a set of annotations applied to a collection of images;   a means for deploying a plurality of deep learning models from the set of applied annotations, the plurality of deep learning models including a nested set of annotations such that a first model segments a first set of features within the image and a second model segments the first set of features of the first model and at least one additional feature of the image; and   a means for transfer learning, wherein at least one of the plurality of deep learning models deployed for a first collection of images is used to accelerate retraining of the deep learning model for a second class of images defined by the second collection.   
     
     
         2 . A method for training and deployment of automated image segmentation algorithms, the method comprising:
 curating collections of images and annotating images stored in a database using one or more user interfaces;   providing a structured library of annotations that form a dictionary for defining features within an image;   training a deep learning model according to a set of annotations applied to a collection of images;   deploying a plurality of deep learning models from the set of applied annotations, the plurality of deep learning models including a nested set of annotations such that a first model segments a first set of features within the image and a second model segments the first set of features of the first model and at least one additional feature of the image; and   transfer learning, wherein at least one of the plurality of deep learning models deployed for a first collection of images is used to accelerate retraining of the deep learning model for a second class of images defined by the second collection.   
     
     
         3 . A computer program product for training and deployment of automated image segmentation algorithms, the computer program product comprising:
 a non-transitory computer readable storage medium having computer readable program code embodied in said medium, the computer readable program code comprising:   computer readable program code to curate collections of images and annotating images stored in a database using one or more user interfaces;   computer readable program code to provide a structured library of annotations that form a dictionary for defining features within an image;   computer readable program code to train a deep learning model according to a set of annotations applied to a collection of images;   computer readable program code to deploy a plurality of deep learning models from the set of applied annotations, the plurality of deep learning models including a nested set of annotations such that a first model segments a first set of features within the image and a second model segments the first set of features of the first model and at least one additional feature of the image; and   computer readable program code to transfer learn, wherein at least one of the plurality of deep learning models deployed for a first collection of images is used to accelerate retraining of the deep learning model for a second class of images defined by the second collection.